2 * linux/fs/file_table.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
8 #include <linux/string.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/init.h>
12 #include <linux/module.h>
14 #include <linux/security.h>
15 #include <linux/eventpoll.h>
16 #include <linux/rcupdate.h>
17 #include <linux/mount.h>
18 #include <linux/capability.h>
19 #include <linux/cdev.h>
20 #include <linux/fsnotify.h>
21 #include <linux/sysctl.h>
22 #include <linux/percpu_counter.h>
24 #include <asm/atomic.h>
26 /* sysctl tunables... */
27 struct files_stat_struct files_stat = {
31 /* public. Not pretty! */
32 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
34 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
36 static inline void file_free_rcu(struct rcu_head *head)
38 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
39 kmem_cache_free(filp_cachep, f);
42 static inline void file_free(struct file *f)
44 percpu_counter_dec(&nr_files);
45 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
49 * Return the total number of open files in the system
51 static int get_nr_files(void)
53 return percpu_counter_read_positive(&nr_files);
57 * Return the maximum number of open files in the system
59 int get_max_files(void)
61 return files_stat.max_files;
63 EXPORT_SYMBOL_GPL(get_max_files);
66 * Handle nr_files sysctl
68 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
69 int proc_nr_files(ctl_table *table, int write, struct file *filp,
70 void __user *buffer, size_t *lenp, loff_t *ppos)
72 files_stat.nr_files = get_nr_files();
73 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
76 int proc_nr_files(ctl_table *table, int write, struct file *filp,
77 void __user *buffer, size_t *lenp, loff_t *ppos)
83 /* Find an unused file structure and return a pointer to it.
84 * Returns NULL, if there are no more free file structures or
85 * we run out of memory.
87 * Be very careful using this. You are responsible for
88 * getting write access to any mount that you might assign
89 * to this filp, if it is opened for write. If this is not
90 * done, you will imbalance int the mount's writer count
91 * and a warning at __fput() time.
93 struct file *get_empty_filp(void)
95 struct task_struct *tsk;
100 * Privileged users can go above max_files
102 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
104 * percpu_counters are inaccurate. Do an expensive check before
107 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
111 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
115 percpu_counter_inc(&nr_files);
116 if (security_file_alloc(f))
120 INIT_LIST_HEAD(&f->f_u.fu_list);
121 atomic_set(&f->f_count, 1);
122 rwlock_init(&f->f_owner.lock);
123 f->f_uid = tsk->fsuid;
124 f->f_gid = tsk->fsgid;
125 eventpoll_init_file(f);
126 /* f->f_version: 0 */
130 /* Ran out of filps - report that */
131 if (get_nr_files() > old_max) {
132 printk(KERN_INFO "VFS: file-max limit %d reached\n",
134 old_max = get_nr_files();
144 EXPORT_SYMBOL(get_empty_filp);
147 * alloc_file - allocate and initialize a 'struct file'
148 * @mnt: the vfsmount on which the file will reside
149 * @dentry: the dentry representing the new file
150 * @mode: the mode with which the new file will be opened
151 * @fop: the 'struct file_operations' for the new file
153 * Use this instead of get_empty_filp() to get a new
154 * 'struct file'. Do so because of the same initialization
155 * pitfalls reasons listed for init_file(). This is a
156 * preferred interface to using init_file().
158 * If all the callers of init_file() are eliminated, its
159 * code should be moved into this function.
161 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
162 mode_t mode, const struct file_operations *fop)
167 file = get_empty_filp();
171 init_file(file, mnt, dentry, mode, fop);
174 EXPORT_SYMBOL(alloc_file);
177 * init_file - initialize a 'struct file'
178 * @file: the already allocated 'struct file' to initialized
179 * @mnt: the vfsmount on which the file resides
180 * @dentry: the dentry representing this file
181 * @mode: the mode the file is opened with
182 * @fop: the 'struct file_operations' for this file
184 * Use this instead of setting the members directly. Doing so
185 * avoids making mistakes like forgetting the mntget() or
186 * forgetting to take a write on the mnt.
188 * Note: This is a crappy interface. It is here to make
189 * merging with the existing users of get_empty_filp()
190 * who have complex failure logic easier. All users
191 * of this should be moving to alloc_file().
193 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
194 mode_t mode, const struct file_operations *fop)
197 file->f_path.dentry = dentry;
198 file->f_path.mnt = mntget(mnt);
199 file->f_mapping = dentry->d_inode->i_mapping;
204 EXPORT_SYMBOL(init_file);
206 void fput(struct file *file)
208 if (atomic_dec_and_test(&file->f_count))
214 /* __fput is called from task context when aio completion releases the last
215 * last use of a struct file *. Do not use otherwise.
217 void __fput(struct file *file)
219 struct dentry *dentry = file->f_path.dentry;
220 struct vfsmount *mnt = file->f_path.mnt;
221 struct inode *inode = dentry->d_inode;
225 fsnotify_close(file);
227 * The function eventpoll_release() should be the first called
228 * in the file cleanup chain.
230 eventpoll_release(file);
231 locks_remove_flock(file);
233 if (file->f_op && file->f_op->release)
234 file->f_op->release(inode, file);
235 security_file_free(file);
236 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
237 cdev_put(inode->i_cdev);
238 fops_put(file->f_op);
239 if (file->f_mode & FMODE_WRITE)
240 put_write_access(inode);
241 put_pid(file->f_owner.pid);
243 file->f_path.dentry = NULL;
244 file->f_path.mnt = NULL;
250 struct file *fget(unsigned int fd)
253 struct files_struct *files = current->files;
256 file = fcheck_files(files, fd);
258 if (!atomic_inc_not_zero(&file->f_count)) {
259 /* File object ref couldn't be taken */
272 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
273 * You can use this only if it is guranteed that the current task already
274 * holds a refcnt to that file. That check has to be done at fget() only
275 * and a flag is returned to be passed to the corresponding fput_light().
276 * There must not be a cloning between an fget_light/fput_light pair.
278 struct file *fget_light(unsigned int fd, int *fput_needed)
281 struct files_struct *files = current->files;
284 if (likely((atomic_read(&files->count) == 1))) {
285 file = fcheck_files(files, fd);
288 file = fcheck_files(files, fd);
290 if (atomic_inc_not_zero(&file->f_count))
293 /* Didn't get the reference, someone's freed */
303 void put_filp(struct file *file)
305 if (atomic_dec_and_test(&file->f_count)) {
306 security_file_free(file);
312 void file_move(struct file *file, struct list_head *list)
317 list_move(&file->f_u.fu_list, list);
321 void file_kill(struct file *file)
323 if (!list_empty(&file->f_u.fu_list)) {
325 list_del_init(&file->f_u.fu_list);
330 int fs_may_remount_ro(struct super_block *sb)
334 /* Check that no files are currently opened for writing. */
336 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
337 struct inode *inode = file->f_path.dentry->d_inode;
339 /* File with pending delete? */
340 if (inode->i_nlink == 0)
343 /* Writeable file? */
344 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
348 return 1; /* Tis' cool bro. */
354 void __init files_init(unsigned long mempages)
357 /* One file with associated inode and dcache is very roughly 1K.
358 * Per default don't use more than 10% of our memory for files.
361 n = (mempages * (PAGE_SIZE / 1024)) / 10;
362 files_stat.max_files = n;
363 if (files_stat.max_files < NR_FILE)
364 files_stat.max_files = NR_FILE;
366 percpu_counter_init(&nr_files, 0);